⚡ Lambda Cloud CLI

A fast, interactive CLI + MCP server for managing Lambda Cloud GPU instances.

Built with Ink (React TUI) · Commander · @modelcontextprotocol/sdk

Install

npm install
npm run build
npm link    # creates global 'lambda' and 'lambda-cli' commands

Setup

lambda config set apiKey YOUR_LAMBDA_API_KEY
lambda config set defaultSshKey YOUR_SSH_KEY_NAME
lambda config set defaultRegion us-east-3

Commands

Examples

# Check GH200 availability
lambda types gh200

# Poll for a GH200 and auto-launch when available
lambda poll gpu_1x_gh200 --auto-launch

# List running instances
lambda instances

# SSH into an instance
lambda ssh <instance-id>

# Push training data
lambda push <instance-id> ./data /home/ubuntu/data

# Terminate all instances
lambda terminate --all

Cost Tracking

The CLI automatically tracks instance uptime and calculates running costs:

• On launch: records timestamp and hourly rate locally
• On lambda instances: shows Uptime and Cost columns + session total
• On lambda terminate: shows final session cost per instance

$ lambda instances
  Status      Name          IP              GPU                    Region       $/hr    Uptime     Cost
  🟢 active   datagen-test  192.222.56.129  1x GH200 480GB        us-east-3   $1.99    5h 14m   $10.44
  💰 Session total: $10.44 across 1 instance(s)

Cost data is stored locally in ~/.config/lambda-cli/cost-tracker.json.

Config Keys

MCP Server (AI Agent Integration)

Lambda Cloud CLI includes a built-in MCP server so AI agents can manage GPU instances programmatically.

Setup for Claude Code / Cursor / Codex

Add to your MCP config (claude_desktop_config.json, .cursor/mcp.json, etc.):

{
  "mcpServers": {
    "lambda": {
      "command": "lambda-cli",
      "args": ["mcp"]
    }
  }
}

Or via npx (no global install):

{
  "mcpServers": {
    "lambda": {
      "command": "npx",
      "args": ["-y", "lambda-cloud-cli", "mcp"]
    }
  }
}

Available Tools

What agents can do

"Launch a GH200, upload my training data, start the fine-tuning job, monitor it, pull the results, and terminate the instance."

All autonomously, with full visibility into costs and availability.

License

MIT